Key takeaways
- GA4 now has a native AI Assistants channel, but to see every ChatGPT, Perplexity and Gemini visit and tie it to leads you need a custom setup.
- Since 2026 GA4 auto-groups ChatGPT, Gemini, Copilot, Grok and Deepseek as AI Assistants, but Perplexity and Claude still fall into Referral.
- Google's own AI Overviews and AI Mode are counted as organic search, not AI, so you infer those from Search Console rather than a GA4 referral.
- A custom channel group placed above Referral, built on a source regex, pulls every AI tool into one line you can report on, retroactively.
- The point is not the headline number. Build an exploration that ties AI sessions to key events, then feed what converts back into your content.
To see AI traffic in GA4, use the native AI Assistants channel for ChatGPT, Gemini, Copilot, Grok and Deepseek, then build a custom channel group that also catches Perplexity and Claude and sits above Referral. Google’s own AI Overviews and AI Mode are counted as organic search, so you track those through Search Console, not a referral.
That is the short version. The longer version matters, because the default setup quietly scatters your AI traffic across three different channels, and a good chunk of it ends up somewhere you would never think to look. Here is how to pull it back together, see which assistants send you real visitors, and tie those visits to enquiries.
Why track AI traffic as its own channel?
Because it is growing fast, it behaves differently from ordinary search, and while it sits hidden inside Referral and Direct you cannot tell whether it earns you anything.
AI-referred traffic to US retail sites grew 138% year on year in May 2026, according to Adobe Analytics. A Leicester plumber is not a US retailer, so take the exact figure with a pinch of salt, but the direction of travel is the same wherever you look. It is also no longer a one-tool story. ChatGPT’s share of generative AI website visits slipped from around 76% to roughly 53% in the year to May 2026 (Similarweb), as Gemini, Perplexity and the rest took ground. So “a bit of ChatGPT traffic” is now several assistants at once, each sending people who found you through an answer rather than a ranked link.
This is a different question from whether AI names you in the first place. We covered that side, the brand mentions, in how to check whether AI recommends your brand. This article picks up after the recommendation lands: the person clicks, arrives on your site, and you want to see it, count it, and connect it to a lead.
Where AI traffic lands in GA4 by default
GA4 now has a native AI Assistants channel, but it does not catch everything, and Google’s own AI answers are not in it at all.
Since 2026, Google’s default channel group has included an AI Assistants channel. It groups any session whose medium is ai-assistant or whose referrer matches a recognised list: ChatGPT, Gemini, Deepseek, Copilot and Grok. That part is genuinely helpful and needs no setup from you.
The gaps are where it gets messy. Perplexity and Claude are not on Google’s list, so their visits drop into plain Referral alongside every other site that happens to link to you. And Google’s own AI Overviews and AI Mode are, by Google’s own definition, excluded from the channel entirely: those clicks are counted as organic Google search, because they begin inside a Google results page. So before you have touched a single setting, your AI traffic is already split three ways.
| AI source | Typical referrer | Default GA4 channel | What you need to do |
|---|---|---|---|
| ChatGPT | chatgpt.com, chat.openai.com | AI Assistants | Nothing, but add to a custom group for reporting |
| Gemini | gemini.google.com | AI Assistants | As above |
| Copilot | copilot.microsoft.com | AI Assistants | As above |
| Grok | grok.com, x.com | AI Assistants | As above |
| Perplexity | perplexity.ai | Referral | Add it to a custom channel group |
| Claude | claude.ai | Referral | Add it to a custom channel group |
| AI Overviews, AI Mode | google / organic | Organic Search | Cannot be split out: infer from Search Console |
Read the last column and the job is clear. Four assistants are handled for you, two need rescuing from Referral, and Google’s own AI cannot be isolated in GA4 at all.
Build a custom channel group for every AI tool
Create one custom channel group with a single AI channel, define it by a source regex plus the ai-assistant medium, and drag it above Referral so it claims the traffic first.
The path is Admin, then Data display, then Channel groups, then Create new channel group, which starts as a copy of the default. Add a new channel, call it something plain like “AI Assistants (all)”, and give it this condition on Source using the “matches regex” operator:
chatgpt\.com|chat\.openai\.com|perplexity\.ai|gemini\.google\.com|claude\.ai|copilot\.microsoft\.com|grok\.com|deepseek\.com
Then add an OR condition so Medium exactly matches ai-assistant, which sweeps in anything Google has already tagged for you. Add new assistants to the regex as they appear, and keep the escaped full stops, because an unescaped dot matches any character.
Ordering is the step people get wrong. GA4 uses first-match logic, so a session lands in the first channel whose rule it fits. Leave your AI channel below Referral and Referral claims chatgpt.com and perplexity.ai before your rule ever runs, and your new channel stays empty while you wonder why. Drag it above Referral and the problem disappears. Two more things worth knowing: a standard GA4 property gives you two custom channel groups to work with, and channel groups apply retroactively, so you see historical AI sessions the moment you save rather than starting a fresh count from today.
The Direct traffic blind spot
A real share of AI traffic arrives with no referrer at all and lands in Direct, where no channel rule can reach it.
Many AI tools, and their mobile apps in particular, strip or simply do not pass a referrer. The visit is as real as any other, but it reaches GA4 looking like someone typing your web address straight into the bar, so it sits in Direct and your tidy AI channel never sees it. There is no clean fix for traffic that arrives genuinely unlabelled, and anyone promising you perfect AI attribution is overselling it.
What you can do is watch for the fingerprint. Keep an eye on Direct sessions landing on deep, specific pages that nobody would ever bookmark or type from memory, especially when they rise in step with your named AI channel. It is a strong hint that your AI traffic is larger than the labelled figure. Treat the named channel as a floor, not a ceiling.
Tie AI visits to actual leads with an exploration
A channel line tells you volume. An exploration tells you whether those visits do anything, by crossing your AI sources with your key events.
Start with the outcomes. Make sure your key events are marked: the enquiries, calls, bookings and form submissions that count as real business. If that groundwork is shaky, our GA4 setup guide walks through marking key events and getting the UK consent layer right first, which is worth doing before you trust any number.
Then build a free-form exploration:
- Open Explore and start a blank Free-form exploration.
- Add your custom channel group (or Session source / medium) as the row dimension, and Landing page as a second dimension beneath it.
- Bring in Sessions, Engaged sessions and your Key events as values.
- Add a segment that isolates AI sources, built on the same regex, so the whole report shows nothing but AI traffic.
Now you can read the part that matters: which assistants send people, which pages they land on, how engaged those sessions are, and how many turn into a key event. That is the line from “AI sent someone” to “AI sent someone who enquired”, and it is the whole reason for doing any of this. Check it monthly rather than daily, because the volumes are small enough that a single day is mostly noise.
What to do with what you find
Use it to decide where to put your effort: lean into the pages and questions AI already sends you to, and fix the ones it should send to but does not.
A handful of moves come up again and again:
- Follow the landing pages. If one page pulls most of your AI visits, it is answering a question cleanly. Study what makes it work, then write more pages like it.
- Match content to real questions. AI sends people who asked something specific. Shape pages that answer those questions directly and near the top, which is the core of answer engine optimisation.
- Judge by conversion, not volume. A small, high-converting stream of AI visits can be worth more than a big, idle one, so read your exploration on key events, not sessions alone.
- Pair it with Search Console. Because AI Overviews hide inside organic, cross-reference your GA4 view with Search Console query trends to work out where Google’s own AI is quietly involved.
This is the join between AI search and measurement, and it is the work we do week in, week out for UK businesses through our analytics and measurement service. Getting found in AI is only half the job. Proving it sent you customers is the other half.
See your AI traffic clearly
Most UK SMBs we look at have AI traffic already. They just cannot see it, because it is scattered across Referral, Direct and organic, and nobody has stitched it back together. Pulling it into one channel and tying it to leads takes an afternoon and changes what you are able to decide.
If you would rather have it set up and read for you, book a free growth review. We will show you what AI is already sending you, whether it converts, and the gaps worth closing first. No lock-in, and if you want us to keep going, the work is scoped monthly with weekly updates. Prefer to talk it through first? Get in touch.